README.md

metadata

license: mit
task_categories:
  - text-to-speech
language:
  - sw

Swahili Speech-to-Text Dataset

This dataset contains paired audio and text data for training and evaluating speech-to-text models in Swahili. The audio files have been processed to remove silence, converted to 44.1kHz mono FLAC format, and are paired with corresponding transcriptions.

Structure

• audio_*.flac: Audio files in FLAC format, named by their corresponding text corpus ID.
• metadata.jsonl: JSON Lines file with metadata for each audio-text pair. Each line is a JSON object with the following fields:
  • id: Integer, the text corpus ID
  • file_name: String, the filename of the audio file
  • text: String, the transcription
  • speaker_id: Integer, the speaker identifier (currently only 1 and 2)

Example metadata entry

{
  "id": 1,
  "file_name": "audio_1.flac",
  "speaker_id": 1,
  "text": "Habari za asubuhi?"
}

Usage

You can load this dataset using the Hugging Face Datasets library:

from datasets import load_dataset

dataset = load_dataset('path/to/hf-dataset', split='train')